The ability to validate credit cards using a mobile device is fruitful for many e-commerce platforms including Uber. Not only does this provide a seamless experience to users, but it also enables the company to verify that a user has physical possession of the credit card. In this talk we will discuss our new application that uses object detection neural networks to scan credit cards.
Traditionally, machine learning models are hosted server side, but with challenges including high bandwidth inputs, low network speeds, and a greater focus on user privacy, hosting these models server side is not always feasible. Recent advancements bring up the possibility of deploying these models directly on the mobile devices. We will discuss the challenges we faced in designing the vision model to run on a mobile device. These include reducing the model’s size footprint and optimizing the model to run on various different types of mobile hardware.